Daniel sent us this one — and it's one of those prompts where the question itself is a window into something much bigger. He lives in central Jerusalem, and from his balcony he can see several of these so-called ghost towers — massive residential skyscrapers where, at night, only a handful of windows are lit. He's not just guessing they're empty. He's looking at them. And what he's asking is: has anyone actually developed a rigorous methodology for estimating occupancy by counting lights? Like, is this a real thing that urban researchers do, or is it just a guy on his balcony squinting at the skyline? And if it is real, how do they do it? What's the protocol?
It is absolutely a real thing, and it's got a surprisingly formal academic pedigree. There's a 2022 paper in the Journal of Urban Technology titled "Light-based Occupancy Estimation in High-Rise Residential Buildings" that lays out the whole framework. This isn't some guy with binoculars — it's a published methodology with error bounds, correction factors, the works.
"Light-based occupancy estimation." They really dressed it up.
They dressed it up, but the core insight is exactly what the prompt describes. You take a photograph of a building at night, you count the lit windows, you divide by the total number of windows, and you apply some adjustments. That's it. The elegance is that it's so simple anyone can do it, but the rigor comes from how you structure the observation.
Where do we even start with this? Because the obvious pushback is: someone could be home with the lights off. Someone could be on vacation with the lights on a timer. You could have cleaning staff in a vacant unit. How do you go from "I counted seventeen lit windows" to "this building is sixty percent empty"?
That's where the methodology gets genuinely clever. Researchers don't just do a single count. The standard protocol uses multiple passes — typically three time windows on each observation night. You go out at eight PM, again at ten PM, and again at midnight. The eight PM window captures people who are home and active. Ten PM captures the evening wind-down. Midnight filters for people who are actually sleeping there — it eliminates a lot of the dinner-party crowd, the people who are in the unit but don't live there.
Because at midnight, if the lights are on, someone's probably brushing their teeth or reading in bed. That's a resident.
And then you repeat this on different days — typically two weekdays and one weekend night. The weekend weighting is important because families are more likely to be home on a Saturday evening than a Tuesday. If a unit is dark on a Tuesday and dark on a Saturday, that's a much stronger signal that it's unoccupied.
You're building a composite picture. It's not "I looked once and counted." It's "I looked nine times across three nights and here's the pattern.
And then you calculate what the literature calls the "lit window fraction" — that's the ratio of illuminated windows to total windows. But you don't just take the raw ratio. You have to adjust for building geometry, because corner units often have windows on two facades and can be double-counted if you're not careful. You have to account for curtain opacity — heavy blackout curtains can mask occupancy. And you have to account for floor height, because higher floors tend to have lower occupancy in these investor-driven buildings anyway.
Wait, say more about that last one. Higher floors are emptier?
In investor-driven luxury towers, the premium units — the penthouses and upper floors — are the ones most likely to be bought as pure investments. They're trophy assets. You park five million dollars in a penthouse and never set foot in it. So when researchers do light counts, they often find that occupancy drops off as you go up. The lower and mid-level floors might show forty or fifty percent lit windows, and the top floors show ten percent.
That's the opposite of how people imagine skyscrapers. You picture the penthouse as the bustling center of some oligarch's social life, but it's actually a concrete vault for capital.
A concrete vault for capital. I'm using that.
You've got your lit window fraction. What's the threshold where researchers say: this building is a problem?
There's a widely-used benchmark in the literature. A building is classified as "underoccupied" or "functionally vacant" if its lit window fraction falls below thirty percent on three separate sampling nights. Below thirty percent, you're not talking about normal vacancy — you're talking about a building that primarily exists as an investment vehicle rather than a place where people live.
And the prompt mentions estimating some Jerusalem towers at ten to twenty percent. So those wouldn't even be borderline cases. They'd be deep in the red zone.
Deep in the red zone. And that aligns with what actual researchers have found. There was a 2024 study out of Hebrew University — this is right in the prompt's backyard — that used satellite nightlight imagery to map occupancy across forty Jerusalem towers. They found a median lit window fraction of eighteen percent for buildings completed after 2020.
So half the buildings were below eighteen percent.
And these are buildings marketed as sold out, or ninety percent sold. The gap between "units sold" and "units lived in" is the entire story here. The sale happened. The money moved. But no one's home.
That's the thing official data misses entirely. The city might report that housing stock increased by two thousand units this year, and on paper that looks like progress. But if seventy percent of those units are dark every night, you haven't actually housed anyone.
This is exactly why light counting has become such an important tool — because official occupancy data is either proprietary or useless. Those are private. Aggregated at the census tract level, so you can't see individual buildings. The postal service knows which addresses receive mail, but that data isn't public. So researchers and journalists are left squinting at the skyline because it's literally the only way to get building-level occupancy data without a subpoena.
It's a kind of information asymmetry. The developer knows exactly how many units are occupied because they manage the building. The city might not even ask. And the public is just supposed to take the "ninety percent sold" press release at face value.
Sold doesn't mean occupied. Sold means a transaction occurred. The buyer could be a shell company in the British Virgin Islands. The unit could sit empty for five years while it appreciates. That's not housing policy — that's a wealth storage system that happens to be shaped like an apartment building.
We've established that light counting is a real, documented technique. But how do you actually do it without a research grant? Let's walk through the methodology step by step. What's the simplest version that still produces useful data?
The simplest version that's still defensible is what I'd call the three-by-three rule. Three nights — two weekdays, one weekend. Three time windows per night — eight PM, ten PM, and midnight. You photograph the same facade from the same position each time. Count the lit windows. Count the total windows. That's your raw lit window fraction.
Then you apply a correction factor, because some people are home with the lights off.
The standard correction factor in the literature is zero point eight five. It accounts for what researchers estimate is about fifteen percent of occupied units where residents are present but not producing visible light. They might be sleeping. They might be watching TV in a dark room. They might be in a room that doesn't face the street. The zero point eight five multiplier says: whatever your raw count is, the actual occupancy is probably about fifteen percent higher.
If I count twenty lit windows out of a hundred, my raw fraction is twenty percent. Apply the correction factor — divide by zero point eight five — and my estimated occupancy is about twenty-three and a half percent.
And that's conservative. Some researchers use zero point eight. Some use zero point nine depending on the building type and local habits. But zero point eight five is the consensus middle ground.
What about ground-floor commercial spaces? A lot of these towers have retail at the base.
You exclude them. The methodology is designed for residential units only. Commercial spaces have different lighting patterns — cleaning crews at night, security lighting, twenty-four-hour businesses. They'd contaminate the data. Most protocols count only floors two and above, or they mask out the ground floor entirely.
What about common areas? Hallway lights, stairwells, lobbies?
You're counting only individual unit windows. This requires some judgment — you have to be able to distinguish a unit window from a hallway window. In most modern towers, the hallway windows are visibly different — they're smaller, or they're frosted, or they're positioned differently on the facade. If you can't distinguish them, you note that as a limitation in your methodology.
Far this is all doable from a balcony with a decent camera. But the prompt also asks about the people who've actually used this approach in the wild. What are the case studies?
The most famous one — and I use "famous" in the niche world of urban occupancy studies — is the 2021 citizen audit of the Living Shangri-La tower in Vancouver. It's the tallest building in the city, sixty-two stories, luxury condos. The developer claimed eighty-five percent sold. A group of housing activists did a systematic light count over multiple nights and published their results: they estimated twenty-two percent occupancy.
Twenty-two percent versus eighty-five percent sold. That's not a rounding error. That's a completely different building.
It triggered a municipal audit. The city of Vancouver ended up investigating foreign ownership and empty homes in the building, and the Shangri-La became the poster child for Vancouver's empty homes tax. All because someone stood on the street and counted lights.
"Someone stood on the street and counted lights" — there's something almost absurdly simple about that. It's the kind of thing that makes you wonder why no one did it sooner.
Because we've been trained to trust official data. If the developer says eighty-five percent sold, and the city doesn't dispute it, and the tax records are private, who's going to stand on the sidewalk and count windows? It feels almost rude. Like you're questioning something you're supposed to take on faith.
Like adopting a feral cat.
I'm not sure that's the analogy, but I appreciate the attempt.
The point is, there's a social norm against checking. And these methodologies exist partly to give people permission to check — to say, no, this isn't nosy, this is citizen science, here's the protocol, here's the error bound, here's the published paper.
And it's not just Vancouver. There was a 2023 study of luxury towers in Miami's Brickell district that used drone-mounted cameras. They flew drones at nine PM on three consecutive Tuesdays and photographed the facades of twelve buildings marketed as luxury residences. They found an average occupancy of thirty-eight percent in buildings that developers claimed were ninety percent sold or more.
So now we're not just standing on the sidewalk. We're in the airspace.
Which brings us to an important methodological refinement — and an ethical boundary. The Miami study used drones because many of the towers had irregular shapes and you couldn't see all facades from street level. The drone gave them a complete picture. But a drone hovering outside someone's window at nine PM is... let's call it a different category of observation than counting lights from a public street.
Yeah, that's where we need to draw the line. And the prompt acknowledges this — it says "this might sound invasive and slightly creepy." There's a difference between aggregation and identification. Counting lit windows across an entire facade is aggregation. You're not identifying which unit is occupied by whom. You're just counting. But if you're flying a drone close enough to see into someone's living room, that crosses into surveillance.
The ethical consensus in the research community is pretty clear on this. Light counting from a public vantage point — a street, a park, your own balcony — is legal and non-invasive. You're observing what's visible to any passerby. The moment you use a telephoto lens to peer into a specific unit, or you fly a drone close enough to see interior details, you've crossed the line from urban research into voyeurism. The distinction is whether you're counting or identifying.
"Counting or identifying." That's the fence.
That's the fence. And it's a sturdy one. When you're counting lit windows on a facade, you don't know anything about who's inside. You don't know their names, their faces, their lives. You just know that at ten PM on a Tuesday, unit fourteen-oh-three had a light on. That's data, not surveillance.
It's worth saying: developers photograph their own buildings constantly. They publish glamour shots of the facade. They're not worried about privacy when they're marketing penthouses. The concern only seems to arise when someone else does the looking.
The selective privacy concern. "Our drone shots are marketing. Your drone shots are stalking.
Covering the covers.
Light counting works, but it's not the only game in town. Let's compare it to the other methods researchers use to measure occupancy — and see where each falls short.
What else is in the toolkit?
The gold standard would be utility meter data. If you could see the electricity consumption of every unit, you'd know with near certainty which ones are occupied. BC Hydro — that's the utility in British Columbia — actually did this for Vancouver's empty homes audit. They cross-referenced meter data with property records and found thousands of units with negligible electricity use over a full year. But that data is almost never public. It's protected by privacy laws, and utilities are reluctant to share it even with city governments without a legal mandate.
It's the best data, and it's locked in a vault.
Locked in a vault. Another method is thermal imaging from satellites. Landsat 8 has a thermal infrared band that can detect heat signatures from buildings. In theory, an occupied unit should be warmer than an empty one — body heat, heating systems, cooking. But the resolution is a hundred meters per pixel. That's useless for individual buildings, let alone individual units. You can see that a neighborhood is warm, but you can't see which windows are lit.
Satellite thermal is like trying to read a book from the International Space Station. You can tell the book exists.
Then there's smart meter data, which is granular down to the minute, but it's privacy-constrained and proprietary. There are postal return rates — you send mail to every unit and see how much comes back as undeliverable. That's a lagging indicator and it only catches units where no one has set up mail forwarding. There's water usage data, which is even more private than electricity. And there's the doorknocker method — literally knocking on doors and asking if someone lives there.
Which has its own biases. People who don't want to be found don't answer doors.
In luxury towers with concierge desks, you're not getting past the lobby anyway. So light counting sits in this sweet spot: it's non-invasive, it requires no special access, it can be done from public space, and it produces data at the individual building level. It's the least bad option in a field where all the good options are locked away.
That's the thesis of citizen science in a nutshell. "The least bad option that's actually available to you.
Sometimes that's enough. The Vancouver Living Shangri-La audit wasn't perfect. They probably missed some occupied units. Their correction factor might have been off. But "approximately twenty-two percent" versus "eighty-five percent sold" — even with a generous margin of error, those numbers are not in the same universe. The methodology doesn't need to be flawless to expose the gap.
Let's talk about the Hebrew University study from 2024. You mentioned it used satellite nightlight imagery to map forty Jerusalem towers. How does that work at the satellite level if Landsat's resolution is a hundred meters?
They didn't use Landsat. They used commercial satellite imagery — the kind from companies like Planet or Maxar — which can now achieve resolutions of thirty centimeters for nightlight data. That's sharp enough to distinguish individual windows. It's not publicly available for free, but it's accessible to researchers with grant funding. And it's getting cheaper every year.
So from space, you can see whether my bedroom light is on.
In theory, yes. In practice, there are atmospheric distortions and angle issues. But the trend line is clear: satellite imagery is approaching the resolution where automated light counting from orbit becomes feasible at city scale. The Hebrew University study found that the median lit window fraction for Jerusalem towers built after 2020 was eighteen percent. That's across forty buildings. And the distribution was heavily skewed — a few buildings were at forty or fifty percent, but the long tail was down near single digits.
Eighteen percent median. That means for every tower at forty percent, there's one at basically zero.
Those near-zero towers are the pure investment plays. They're not even trying to rent them out. They're just holding the asset. The lights are off because no one has ever turned them on.
That's a chilling image. A twenty-story building where the lights have never been turned on.
The prompt mentions seeing this from the balcony — that direct observation of a massive structure with maybe ten or fifteen lit windows. What's powerful about that is that it cuts through the abstraction. You can read about vacancy rates in a report, but when you're standing there looking at it, the gap between what was promised and what exists is undeniable.
There's something about direct observation that official statistics can't replicate. A census tract number says "this area has a vacancy rate of X." But your eyes tell you: that building is empty. That building is a ghost. And the prompt's instinct — that even a single photograph could estimate occupancy — is basically correct. One good nighttime photo of a facade gives you a rough estimate. Five observations at different times gives you a defensible one.
That's exactly what the prompt proposes: five estimates at different times on different days. That maps almost perfectly to the published protocols. The standard is three nights with three time windows each — nine observations. Five is in the same ballpark, especially if you're weighting the weekend observations higher and excluding commercial ground-floor units. The prompt's intuition about methodology is basically what the researchers landed on after years of refinement.
Which is satisfying, honestly. It means the guy on the balcony was doing urban science without knowing it.
Citizen science often works that way. Someone notices something, starts counting, and only later discovers there's a whole academic literature that validates their approach. The prompt's estimate of ten to twenty percent occupancy for the Jerusalem towers? That's not a wild guess. That's consistent with what the Hebrew University researchers found using satellite data and what the Vancouver auditors found using street-level light counts. The numbers converge.
If someone listening wants to try this tonight — they've got a visible tower from their window or their street — what's the actual protocol? Walk me through it like I'm about to do it.
Step one: pick your building and your observation point. It needs to be a spot you can return to consistently — same angle, same distance, same framing. Step two: choose your nights. Two weeknights, one weekend night. Avoid holidays, avoid severe weather, avoid nights when there's a big event that might distort occupancy patterns. Step three: on each night, photograph the building at eight PM, ten PM, and midnight. Use the same camera settings if possible. Step four: count the total residential windows on the facade you can see. Exclude ground-floor commercial, exclude common areas. Step five: for each photo, count the lit windows. Step six: calculate your raw lit window fraction — lit windows divided by total windows. Step seven: apply the zero point eight five correction factor. That's your estimated occupancy.
What do you do with that number?
This is where it gets interesting. You can cross-reference your count with publicly available data. Look up the building's address on the local property tax roll — many jurisdictions flag whether a property is owner-occupied or investor-owned. Check AirDNA or similar platforms to see how many units are listed as short-term rentals, because a unit that's on Airbnb isn't occupied by a resident — it's a hotel room. If your light count shows twenty percent occupancy and the tax roll shows eighty percent investor ownership, those numbers tell a consistent story.
You can share your data. The prompt mentions not wanting to start a spreadsheet, but the aggregate power of dozens of people doing this across a city is enormous. If fifty people each audit one tower and publish their methodology and raw counts, you've suddenly got a crowdsourced occupancy map that might be more accurate than anything the city has.
This is already happening in some cities. There are housing advocacy groups that run "balcony audit" projects — they train volunteers on the methodology, assign buildings, and compile the results. The data has been used in city council hearings, in zoning debates, in campaigns for vacancy taxes. It's not just an exercise. It's a policy tool.
The beauty of it is that the methodology is transparent. A developer can dispute your number, but they can't dispute your method without proposing a better one. And if they have better data — if they know the actual occupancy — they're welcome to share it. The light count puts the burden of proof where it belongs.
The light count says: here's what we can see. If you have better information, show it. And if you don't, then our number is the best available estimate.
All of this raises an uncomfortable question, though. At what point does counting lights become surveillance? We touched on drones and telephoto lenses, but even street-level observation has a creep factor if you're doing it systematically. Where's the line?
The line, as I said, is aggregation versus identification. But there's a second dimension: purpose and publication. If you're counting lights to produce an aggregate occupancy statistic for a building, and you publish only the aggregate, that's research. If you're counting lights to figure out whether your ex-girlfriend is home, that's stalking. Same action, different intent, different outcome.
The publication part matters. When the Vancouver group published their Shangri-La data, they didn't say "unit fourteen-oh-three was occupied on Tuesday." They said "we estimate twenty-two percent occupancy across the building." The individual units disappear into the statistic.
And that's also why the methodology doesn't require you to track which specific units are lit across observations. You're not building a pattern-of-life profile for unit fourteen-oh-three. You're just counting total lit windows per observation and averaging. The unit-level data is never collected.
The protocol itself has privacy built in. You're not even tempted to identify individuals because your method doesn't capture individual-level patterns.
And that's not an accident. The researchers who developed these protocols were aware of the privacy concerns and designed the methodology to avoid them. Count, don't track. Aggregate, don't identify. Publish the building, not the unit.
What about the arms race aspect? If light counting becomes widespread — if city governments start using it, if advocacy groups are publishing occupancy maps — won't developers just adapt? Install automated blinds, use smart lighting to simulate occupancy, design facades that make light counting harder?
This is already beginning. There are buildings in Dubai and London where the facade lighting is so elaborate — decorative LED strips, illuminated spandrels, glowing crown features — that you can't distinguish unit windows from architectural lighting. It's the architectural equivalent of signal jamming. You flood the visual channel with noise so the signal — is someone home?
"The architectural equivalent of signal jamming." So the ghost towers are learning to hide their ghostliness.
As satellite resolution improves, we might see a more aggressive version of this. If governments start using automated light counting from satellite imagery to enforce vacancy taxes, developers will respond. Smart home systems that randomize lighting patterns. Curtain automation tied to sunset timers. Facade materials that obscure window illumination from above. The arms race between observers and observed is just getting started.
That's a strange future to imagine. A building that's empty but looks occupied, because the building itself is programmed to perform occupancy.
The performative building. It's already happening with short-term rentals — hosts use smart plugs and automated blinds to make units look lived-in for regulatory inspections. Scaling that to an entire tower is just a matter of budget.
Which brings us back to why the simple, low-tech method matters. A guy on a balcony with a camera and a spreadsheet is harder to fool than a satellite, in some ways. You're closer. You can see details. You can notice that the "occupied" unit on the fourteenth floor has the exact same lighting pattern every night at exactly ten PM — because it's a timer, not a person.
The human observer can detect the uncanny valley of fake occupancy. A satellite just sees photons.
Where does this leave us? What should someone actually do with this information?
I think there are three takeaways. First, light counting is real, it's published, it's methodologically sound within known error bounds, and you can do it tonight with your phone. The three-by-three rule — three nights, three time windows, two weekdays and one weekend — will give you a defensible estimate for any visible building.
Second, cross-reference your count with whatever public data exists. Tax rolls, short-term rental listings, property records. The light count is one data point. Combined with ownership data, it tells a story.
Third, share your methodology and your results. The power of this approach scales with participation. One person's balcony audit is an anecdote. Fifty people's balcony audits, using the same protocol, is a dataset that city councils can't ignore.
The caveat, which we should state clearly: this is a diagnostic tool, not a definitive measure. Use it to ask better questions, not to draw final conclusions. Your light count says "this building appears to be significantly underoccupied." It doesn't say why. It doesn't say who owns the empty units. It doesn't say what should be done about it. It just says: look at this. Something is off here.
That's the role of citizen science in a nutshell. You're not replacing the experts. You're generating the question that forces the experts to pay attention.
Tonight, go look at your skyline. Count the lights. Ask what's really there. The ghost towers are visible if you choose to look.
Now: Hilbert's daily fun fact.
Hilbert: The word "abacus" may trace back to the Hebrew word "abaq," meaning dust — referring to the practice of drawing calculation lines in sand. But in the Azores during the nineteen-hundreds, a distinct variant called the "conta de milho" — literally "corn count" — used dried maize kernels on a grooved wooden board, a method brought by Portuguese settlers who adapted it from earlier Mediterranean counting trays.
Hilberts fun fact somehow involves corn and I'm not sure how to feel about that.
I'm going to call you that from now on.
This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop for that deeply personal fun fact. If you enjoyed this episode, leave us a review — it helps more people find the show. We're at myweirdprompts.com and wherever you get your podcasts. Go count some lights.
